Thought it was cool to see this released today: https://pytorch.org/2018/05/02/road-to-1.0.html
Looks like Facebook is going to be making a big push to get Pytorch production-ready.
4 Likes
Great news! But what do we do before it is released? 
My Process. Train using Fast.AI. Take the PyTorch Model and deploy directly in Flask or you can export to ONNX format and deploy in any other DL Framework. So unless you are expecting to serve thousands of requests per second, you don’t to think about it. If you do serve lots of requests per second, you can hire experts that can help.
5 Likes
Keep on exactly as you are right now. There’s no end-user changes when doing model development. Eventually, you can freeze parts of the network architecture as you start to care more about about performance, mobile applications, etc… But the most impressive part is all the new stuff is optional.